Testing a Cloud AI Agent: From Data Analysis to PPT to Video with a Single Input

Hello, I'm Ai Learning's Lao Zhang.

I have been following the AI Agent space in China and recently tried Skywork's newly revamped platform, which has shifted from generating individual assets to executing entire projects end‑to‑end.

Unlike the cumbersome local OpenClaw setup that requires API configuration, a dedicated host and sufficient compute power, Skywork runs everything in the cloud: you register and start using it instantly, and it continues processing even when your computer is off.

Skywork adds three key capabilities:

24/7 cloud execution – tasks run on the cloud while you sleep or are in meetings.

Parallel multimodal creation – documents, PPTs, images, tables, webpages and videos can be generated simultaneously.

Professional skill collaboration – each modality is backed by an expert‑level agent that can be invoked individually or chained together like LEGO blocks.

Test 1: Data‑analysis report

I exported two Excel files from a WeChat public account – one with 30‑day user growth data and another with article reading statistics.

Skywork provides a suite of domestic flagship large models (Kimi K2.6, DeepSeek V4 Pro, GLM 5.1, MiniMax M2.7, Qwen 3.6 Plus) that can be used without configuring any API.

The recommended choice is the "smart‑selected" model, while smaller tasks can be delegated to lighter models such as DeepSeek V4 Flash or Qwen 3.6 Flash.

The workflow includes skill loading, Jupyter and Bash execution, and final artifact generation. The platform also allows re‑optimizing any output – e.g., polishing, expanding, shortening or changing tone.

Test 2: PPT generation

From the same input box I generated a PPT directly from the data‑analysis report without uploading any file.

The system fetched the HTML source, created an outline, and used a built‑in PPT generation skill to produce the slides.

The resulting PPT can be regenerated at the element level or the whole‑page level, downloaded as a .pptx file or saved as a PDF.

Test 3: Video generation

For video creation Skywork uses its own high‑performance models – Seedance 2.0, Kling 3.0, etc., with Seedance 2.0 being the top performer.

The task involved generating a narration script, producing spoken audio, and stitching everything together with FFmpeg.

The whole process took a few minutes, during which Skywork handled errors automatically.

Test 4: Image generation

Skywork also offers advanced image generation, surpassing my expectations from the previous Image 2 tool.

A hidden feature lets you save prompt templates as presets, which can be reused for image or video generation.

Another discovered feature is automatic skill discovery: frequently used prompts are recognized, a skill is created via the /skill‑creator endpoint, and can be invoked later with @ syntax.

Test 5: Feishu integration and scheduled tasks

Integrating with Feishu is straightforward – just provide the App ID and App Secret.

After linking, I can issue commands from my phone, such as scheduling a daily task that fetches AI industry news at 8 am and sends the compiled summary via Feishu.

Conclusion

Skywork has evolved from a single‑point AI tool into a cloud‑based project execution system. By giving it a complete goal – read exported Excel files, produce a data‑analysis report, turn the report into a PPT, generate narration and audio, synthesize a video, and finally create images – the platform handled every step in the cloud, chaining Jupyter/Bash calls, breaking down tasks, and allowing iterative editing of documents, PPTs, images and videos.

This eliminates the need to set up deployments, configure APIs, or juggle multiple tools, effectively combining a data analyst, PPT designer, video producer, visual designer and automation assistant into one cloud workspace.